submit: publish merged report + gallery in one commit

Replace the per-file gallery-thumbnail uploads (one upload_file commit
per fixture, ~80+ for a full fan-out) with a single create_commit that
stages reports/<id>.{html,json} plus every renders/<id>/<fixture>.png at
once. Collapses finalize from ~86 commits to one publish commit, cutting
post-shard finalize from minutes to well under a minute and removing the
429 "concurrency queue" failures that stranded earlier submissions.

Co-authored-by: Cursor <cursoragent@cursor.com>

submit.py

@@ -105,6 +105,7 @@ import gradio as gr
 from cadgenbench.common.paths import data_inputs_dir
 from cadgenbench.common.validity import parse_step
 from huggingface_hub import (
+    CommitOperationAdd,
     HfApi,
     fetch_job_logs,
     hf_hub_download,
@@ -1295,8 +1296,9 @@ def _merge_shards_and_publish(
         html_path = tmp / f"{submission_id}.html"
         html_path.write_text(html, encoding="utf-8")
 
-        _upload_reports(submission_id, html_path, report_json)
-        _upload_gallery_renders_from_dir(submission_id, merged_run)
+        _publish_reports_and_gallery(
+            submission_id, html_path, report_json, merged_run,
+        )
         return report_json["run_summary"]
     finally:
         shutil.rmtree(tmp, ignore_errors=True)
@@ -1326,83 +1328,70 @@ def _build_report_json(run_dir: Path) -> dict[str, Any]:
     return {"run_summary": summary, "per_fixture_results": per_fixture}
 
 
-def _upload_reports(
+def _publish_reports_and_gallery(
     submission_id: str,
     html_path: Path,
     report_json: dict[str, Any],
+    run_dir: Path,
 ) -> None:
-    """Upload ``reports/<id>.html`` + ``reports/<id>.json`` to the Hub.
-
-    Mirrors ``eval_job.py``'s ``_upload_reports`` so the merged-shard
-    artifacts land at the exact paths the leaderboard + the row-flip
-    expect. Uses the process ``HfApi`` (Space HF_TOKEN env).
+    """Publish report HTML/JSON + every gallery thumbnail in one commit.
+
+    The merged artifacts land at the exact paths the leaderboard + the
+    row-flip expect: ``reports/<id>.{html,json}`` plus one
+    ``renders/<id>/<fixture>.png`` per fixture that produced an ``iso``
+    thumbnail. All of it goes in a single ``create_commit`` rather than
+    one ``upload_file`` per file: a fan-out can stage ~80+ thumbnails,
+    and a commit-per-file both serialises the publish (slow) and hammers
+    the dataset's commit endpoint (the 429 "concurrency queue" failures
+    that stranded earlier runs). One commit is atomic, fast, and
+    rate-limit friendly.
+
+    A fixture with no ``iso.png`` (missing output / render that never
+    ran) is skipped, matching the single-job behaviour; the gallery
+    draws the dashed "invalid" cell from the row, so an absent thumbnail
+    is not an error.
     """
-    _with_hub_retries(
-        lambda: _HF_API.upload_file(
-            path_or_fileobj=str(html_path),
+    operations: list[CommitOperationAdd] = [
+        CommitOperationAdd(
             path_in_repo=f"{REPORTS_DIR}/{submission_id}.html",
-            repo_id=HF_SUBMISSIONS_REPO,
-            repo_type="dataset",
-            commit_message=f"add merged HTML report for {submission_id}",
+            path_or_fileobj=str(html_path),
         ),
-        what="report-html upload",
-    )
-    _with_hub_retries(
-        lambda: _HF_API.upload_file(
+        CommitOperationAdd(
+            path_in_repo=f"{REPORTS_DIR}/{submission_id}.json",
             path_or_fileobj=json.dumps(
                 report_json, ensure_ascii=False, indent=2,
             ).encode("utf-8"),
-            path_in_repo=f"{REPORTS_DIR}/{submission_id}.json",
-            repo_id=HF_SUBMISSIONS_REPO,
-            repo_type="dataset",
-            commit_message=f"add merged JSON report for {submission_id}",
         ),
-        what="report-json upload",
-    )
-    logger.info("Uploaded merged reports/%s.{html,json}", submission_id)
-
-
-def _upload_gallery_renders_from_dir(
-    submission_id: str,
-    run_dir: Path,
-) -> None:
-    """Stage one ``iso`` thumbnail per fixture for the leaderboard gallery.
-
-    Mirrors ``eval_job.py``'s ``_upload_gallery_renders`` but reads from
-    the merged run dir: every ``<run_dir>/<fixture>/renders/iso.png``
-    becomes ``renders/<id>/<fixture>.png``. A fixture with no ``iso.png``
-    (missing output / render that never ran) is skipped, matching the
-    single-job behaviour; the gallery draws the dashed "invalid" cell
-    from the row, so an absent thumbnail is not an error.
-    """
-    staged: list[tuple[Path, str]] = []
+    ]
+    render_count = 0
     for fixture_dir in sorted(d for d in run_dir.iterdir() if d.is_dir()):
         iso_png = fixture_dir / "renders" / f"{GALLERY_THUMB_VIEW}.png"
         if iso_png.is_file():
-            staged.append((iso_png, fixture_dir.name))
-    if not staged:
-        logger.info("No gallery renders to upload for %s", submission_id)
-        return
-    for iso_png, fixture_name in staged:
-        _with_hub_retries(
-            lambda iso_png=iso_png, fixture_name=fixture_name: (
-                _HF_API.upload_file(
-                    path_or_fileobj=str(iso_png),
+            operations.append(
+                CommitOperationAdd(
                     path_in_repo=(
-                        f"{RENDERS_DIR}/{submission_id}/{fixture_name}.png"
-                    ),
-                    repo_id=HF_SUBMISSIONS_REPO,
-                    repo_type="dataset",
-                    commit_message=(
-                        f"add gallery render {fixture_name} for {submission_id}"
+                        f"{RENDERS_DIR}/{submission_id}/{fixture_dir.name}.png"
                     ),
+                    path_or_fileobj=str(iso_png),
                 )
+            )
+            render_count += 1
+
+    _with_hub_retries(
+        lambda: _HF_API.create_commit(
+            repo_id=HF_SUBMISSIONS_REPO,
+            repo_type="dataset",
+            operations=operations,
+            commit_message=(
+                f"publish merged report + {render_count} gallery render(s) "
+                f"for {submission_id}"
             ),
-            what="gallery-render upload",
-        )
+        ),
+        what="merged report+gallery publish",
+    )
     logger.info(
-        "Uploaded %d gallery render(s) under %s/%s/",
-        len(staged), RENDERS_DIR, submission_id,
+        "Published reports/%s.{html,json} + %d gallery render(s) in one commit",
+        submission_id, render_count,
     )